What's new in OpenAI's GPT‑5.4?
Major features and what they change
OpenAI released GPT‑5.4 as a family of models aimed at professional work and agentic workflows. The company rolled out two main variants — a "Thinking" edition focused on deeper reasoning and a higher‑capacity "Pro" tier — and emphasized new abilities that push the model from a conversational assistant toward tools that can operate within users' software environments.
What the update delivers
- Native computer use: the model can perform tasks that interact with files and apps, enabling it to run workflows rather than only return text.
- Much larger context windows: APIs now support context sizes up to about one million tokens, letting the model handle long documents, entire projects, or extended multi‑step interactions without losing track.
- Improved tool calling and integrations: OpenAI rebuilt tool‑calling to make plug‑ins and app integrations more reliable, and it introduced dedicated spreadsheet capabilities for Excel and Google Sheets to speed knowledge‑work tasks.
- Product tiers and pricing: GPT‑5.4 is available in distinct commercial tiers; public reporting has listed per‑token prices for both standard and Pro offerings, reflecting the model’s positioning for enterprise workflows.
Why this matters
- Workflow automation: by adding native computer control and stronger tool use, the model is designed to complete multi‑step business tasks — composing, calculating, and filing results — which lowers the barrier to automating knowledge work.
- Enterprise uptake: large context windows and tighter app integrations make the model more attractive to teams that work with long documents or data‑heavy processes.
- Safety and governance: as models gain the ability to act inside applications, oversight and access controls become more important. Companies will need new guardrails for data, auditing, and error handling.
OpenAI frames GPT‑5.4 as a step toward more autonomous, production‑ready AI agents. The model’s new capabilities could accelerate adoption in offices and developer tooling, but they also raise practical questions about control, accuracy, and how enterprises will govern AI that can operate inside their software stacks.